← 返回 Skills 市场
Paper Deep Reading
作者
c-narcissus
· GitHub ↗
· v1.1.3
· MIT-0
171
总下载
0
收藏
0
当前安装
6
版本数
在 OpenClaw 中安装
/install paper-deep-reading
功能描述
Produce a source-aware, research-generative, MIT-0-compatible single-file deep-reading report for a research paper, with a rich narrative body, a final claim...
安全使用建议
This package appears internally consistent and implements a local, source-aware deep-reading pipeline. Before installing, consider: 1) Review the three scripts (extract_latex_paragraphs.py, render_inline_trace_report.py, validate_traceability.py) if you want to be sure they will only run on inputs you supply — they will read any files in the input directory you point them at. 2) The SKILL.md suggests fetching arXiv/OpenReview sources; the skill itself does not include network-fetch code, so an agent with web access could download external papers when following the instructions — ensure you trust the agent's network capabilities. 3) requirements.txt lists pyyaml and markdown though the scripts use only stdlib; this is harmless but slightly inconsistent. 4) There's a minor metadata/version mismatch between registry metadata (1.1.3) and _meta.json (1.4.1) — not a functional risk but worth noting. 5) License is permissive (MIT-like / MIT-0 intent) but confirm it matches your redistribution needs. If you are concerned about data exfiltration, restrict the agent's network access and only run the skill on inputs you explicitly provide.
功能分析
Type: OpenClaw Skill
Name: paper-deep-reading
Version: 1.1.3
The skill bundle provides a sophisticated pipeline for deep, source-grounded analysis of research papers, focusing on computer science literature. It includes Python scripts (extract_latex_paragraphs.py, render_inline_trace_report.py, and validate_traceability.py) designed to extract structured data from LaTeX sources and maintain a strict claim-to-evidence traceability manifest. The instructions in SKILL.md and the research methodology documents are entirely focused on academic analysis, author-intent reconstruction, and artifact generation, with no evidence of malicious intent, data exfiltration, or unauthorized command execution.
能力标签
能力评估
Purpose & Capability
Name/description, templates, and the three Python scripts all implement a local, source-aware pipeline (extract LaTeX paragraphs, render claim->evidence appendix, validate manifests). No unrelated environment variables, binaries, or cloud credentials are requested. The files and behavior are proportional to a single-file, traceability-focused paper reading skill.
Instruction Scope
SKILL.md instructs assembling the 'best available evidence package' and prefers arXiv/source/OpenReview when available. That implies fetching remote sources if the agent has web access, but the included scripts operate on local files only. The instructions do not instruct reading unrelated system secrets, but they do expect the agent to open user-supplied PDFs/LaTeX trees; if the agent is given broad filesystem or network access, it could read whatever is provided. This is expected for the skill's purpose but worth noting.
Install Mechanism
No install spec; this is an instruction-plus-scripts package. No downloads or archive extraction are present. requirements.txt lists markdown and pyyaml, but the shipped scripts only use Python stdlib; those deps appear optional/unnecessary but low risk.
Credentials
No required env vars, no primary credential, and no config paths requested. The skill does not ask for unrelated secrets or system tokens.
Persistence & Privilege
always:false (default). The skill writes local artifacts (report.md, JSON manifests) as intended. It does not attempt to modify other skills or global agent configuration. Autonomous invocation is allowed by default but is not combined with other concerning privileges.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install paper-deep-reading - 安装完成后，直接呼叫该 Skill 的名称或使用
/paper-deep-reading触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.1.3
**Summary:**
Expanded research-generative capabilities and streamlined package discipline for deep-reading reports; improved appendix placement, artifact structure, and language flexibility.
- Adds research-generative analysis guided by a new methodology and references.
- Moves detailed claim-to-evidence locators to a final appendix, keeping the report body readable.
- Introduces a machine-readable research lens artifact (research_lens.json).
- Explicitly supports ClawHub/OpenClaw packaging and MIT-0 license compliance.
- Tightens package by removing CHANGELOG.md and README.md; now includes LICENSE.txt and relevant templates only.
- Improves language selection: report language matches user request, with consistent technical identifier treatment.
v1.1.2
paper-deep-reading v1.1.2
- Initial release of a traceable deep-reading pipeline for research papers.
- Added scripts for extracting LaTeX paragraphs and validating traceability.
- Introduced templates for artifact indexing and traceability manifests.
- Now outputs both a detailed Markdown report and supporting JSON artifacts for claim-to-evidence mapping.
- Provides reference files for artifact contracts and locator policies.
- Includes a requirements.txt and comprehensive README for setup and usage.
v1.1.1
No changes detected in this version.
- Version number remains 1.1.0 in both current and previous files.
- No file changes detected; instructions and content are unchanged.
- No new features, fixes, or documentation updates.
v1.1.0
**Major update: Now source-aware, with improved evidence-gathering and optional storyboard output.**
- Attempts to gather the best available paper sources, preferring arXiv LaTeX, and clearly reports which sources were used.
- Optionally generates a cartoon storyboard summarizing the main idea, if image generation is available.
- Enhanced handling for ICLR papers: integrates OpenReview reviews and rebuttals when possible.
- Explicitly distinguishes LaTeX-primary, PDF-primary, or mixed-source readings, and transparently documents any mismatches or missing sources.
- Retains comprehensive, Markdown-only deep reading report, but is now more specific about input handling and source reliability.
v1.0.1
**Standalone skill now outputs only Markdown-based deep reading reports, removing dependencies and machine-readable artifacts.**
- Now produces only Markdown reports: no JSON, YAML (except skill file), spreadsheets, code, or manifest outputs.
- Changed focus to standalone use; removed requirements for upstream manifests, bundle contracts, or downstream processing.
- Revised and condensed documentation for clearer instructions and mandatory report content.
- All workflow scripts, JSON/YAML templates, and contract schemas have been removed from the repository.
- Added a Markdown report template to guide consistent, formula-preserving outputs aligned with new requirements.
v1.0.0
Initial release of paper-deep-reading skill
- Deeply reads one or a small batch of papers, producing formula-preserving, figure-aware, reviewer-aware reports and graph-ready artifacts.
- Includes detailed rules for claim audits, formula explanation, theory-practice mapping, module critiques, and more.
- Produces per-paper detailed reports and cross-paper comparison artifacts when needed.
- Ensures all reports meet or exceed benchmark examples from top conferences.
- Integrates reviewer and innovation-mining sections to support downstream graph-building and research direction discovery.
元数据
常见问题
Paper Deep Reading 是什么？
Produce a source-aware, research-generative, MIT-0-compatible single-file deep-reading report for a research paper, with a rich narrative body, a final claim... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 171 次。
如何安装 Paper Deep Reading？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install paper-deep-reading」即可一键安装，无需额外配置。
Paper Deep Reading 是免费的吗？
是的，Paper Deep Reading 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Paper Deep Reading 支持哪些平台？
Paper Deep Reading 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Paper Deep Reading？
由 c-narcissus（@c-narcissus）开发并维护，当前版本 v1.1.3。
推荐 Skills